Fan self-adaptive speed regulation control method for green building

Abstract

The invention relates to the technical field of self-adaptive control, in particular to a fan self-adaptive speed regulation control method for a green building. Collecting indoor and outdoor environment and operation data, and combining building information model data to perform multi-source heterogeneous data fusion to generate a high-dimensional feature vector; establishing a dynamic heat balance model, predicting a future environment state by using an LSTM network, and generating a control reference trajectory; constructing an adaptive control strategy by adopting a DQN network, and generating an initial rotating speed control sequence by taking the high-dimensional feature vector and the control reference trajectory as input; constructing a model prediction control framework, combining the control reference trajectory and the initial rotating speed control sequence, optimizing the rotating speed control sequence through distributed calculation, and generating an optimized rotating speed sequence; adjusting LSTM network parameters and a DQN strategy and detecting abnormity by comparing the deviation between a closed-loop control prediction state and an actual state; according to the invention, the effectiveness of adaptive control can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of adaptive control, and specifically to an adaptive speed regulation control method for a fan in a green building. Background Art

[0002] To achieve effective control of a complex indoor environment system, existing technologies mostly adopt architectures such as model reference adaptive control (MRAC) or self-tuning regulator (STR). These methods identify the dynamic characteristics of the system online and adjust the controller parameters accordingly to achieve the purpose of adaptive control. For example, by updating the process model parameters online to adjust the gains of a PID controller, or adjusting the feedforward and feedback control laws in MRAC to track a reference model.

[0003] Traditional speed regulation control methods mostly adopt fixed speeds or simple PID control, which are difficult to adapt to the dynamic changes in the building interior environment, resulting in poor control effects. Although existing adaptive control systems can partially solve the dynamic regulation problem, most of them rely on single-sensor data and lack the ability of multi-source data fusion and global optimization. In addition, existing technologies rarely combine the real-time energy consumption targets of green buildings with user comfort requirements, limiting the intelligence level of the control system; at the same time, the control systems of existing technologies do not have the ability of adaptive learning and model prediction functions, and it is difficult to achieve the optimal energy saving and comfort balance.

[0004] Therefore, an adaptive speed regulation control method for a fan in a green building is proposed. Summary of the Invention

[0005] The purpose of the present invention is to provide an adaptive speed regulation control method for a fan in a green building. First, collect indoor and outdoor environment and operation data, combine building information model data, and perform multi-source heterogeneous data fusion to generate high-dimensional feature vectors; establish a dynamic thermal balance model, use an LSTM network to predict the future environmental state, and generate a control reference trajectory; use a DQN network to construct an adaptive control strategy, with the high-dimensional feature vector and the control reference trajectory as inputs, to generate an initial speed control sequence; construct a model predictive control framework, combine the control reference trajectory and the initial speed control sequence, and optimize the speed control sequence through distributed computing to generate an optimized speed sequence; by comparing the deviation between the closed-loop control predicted state and the actual state, adjust the LSTM network parameters and DQN strategy and detect anomalies; the present invention can improve the effectiveness of adaptive regulation in the Internet of Things control system.

[0006] To achieve the above purpose, the present invention provides the following technical solutions:

[0007] An adaptive speed regulation control method for a fan in a green building, comprising:

[0008] Collect indoor and outdoor environmental data and the operating status data of the fan through the Internet of Things sensor network. Combine the building information model data, and use the Kalman filtering algorithm to perform multi-source heterogeneous data fusion to generate a high-dimensional feature vector.

[0009] Establish an indoor environmental dynamic thermal balance model, use the LSTM network to process the high-dimensional feature vector, predict the future environmental state, and generate a control reference trajectory.

[0010] Adopt a DQN network to construct an adaptive control strategy. Using the high-dimensional feature vector and the control reference trajectory as inputs, and taking the energy-saving rate, PMV thermal comfort index, and equipment life as multi-objective reward functions, generate an initial rotational speed control sequence.

[0011] Construct a model predictive control framework. Combine the control reference trajectory and the initial rotational speed control sequence, and optimize the rotational speed control sequence through distributed computing to generate an optimized rotational speed sequence.

[0012] Input the optimized rotational speed sequence into the dynamic thermal balance model to generate a closed-loop control prediction state. By comparing the deviation between the closed-loop control prediction state and the actual environmental state, dynamically adjust the LSTM network parameters and DQN strategy, and use the isolation forest algorithm to detect anomalies.

[0013] Further, the indoor and outdoor environmental data includes: temperature, humidity, carbon dioxide concentration, and personnel density; the fan operating status data includes: fan rotational speed, fan power, fan vibration, and fan temperature; the building information model data includes geometric information, material property information, and functional information.

[0014] Further, the process of establishing an indoor environmental dynamic thermal balance model, using the LSTM network to process the high-dimensional feature vector, and predicting the future environmental state to generate a control reference trajectory includes:

[0015] Based on the principles of building thermodynamics, combine the room volume, material thermal conductivity, and duct layout in the building information model data to establish an indoor environmental dynamic thermal balance model for calculating the changes in real-time temperature, humidity, and CO2 concentration.

[0016] Input the high-dimensional feature vector into the LSTM network, process the feature sequence in a preset time window, and obtain the predicted temperature, humidity, and carbon dioxide concentration.

[0017] Use the dynamic thermal balance model to impose physical constraints on the LSTM prediction results, correct the prediction deviation, and generate a control reference trajectory in a time series format.

[0018] Transmit the control reference trajectory back to the edge device from the cloud through the MQTT protocol as the input for adaptive control and model predictive control.

[0019] Furthermore, an adaptive control strategy is constructed using a DQN network. Taking the high-dimensional feature vector and the control reference trajectory as inputs, and taking the energy-saving rate, PMV thermal comfort index, and equipment life as multi-objective reward functions, the process of generating the initial rotational speed control sequence includes:

[0020] Construct a DQN network, define the discrete action space of the fan rotational speed, and use the preprocessed high-dimensional feature vector and the control reference trajectory as inputs;

[0021] Design a multi-objective reward function, taking the energy-saving rate, PMV thermal comfort index, and equipment life as optimization objectives, calculate the reward value, and guide the DQN to learn the optimal control strategy;

[0022] Train the DQN using a greedy strategy in the cloud, store the state-action-reward data using the experience replay mechanism, and update the neural network parameters;

[0023] Perform DQN inference on the edge device, and generate the initial rotational speed control sequence according to the real-time input high-dimensional feature vector and the control reference trajectory.

[0024] Furthermore, construct a model predictive control framework. Combining the control reference trajectory and the initial rotational speed control sequence, the process of optimizing the rotational speed control sequence through distributed computing to generate the optimized rotational speed sequence includes:

[0025] Construct a model predictive control framework, combine the control reference trajectory and the dynamic thermal balance model, and define the prediction horizon, control horizon, and state transition function;

[0026] Taking the initial rotational speed control sequence as the initial value, design an optimization objective function, comprehensively consider the energy-saving rate and PMV thermal comfort index, and set the constraint conditions;

[0027] Decompose the optimization problem into sub-problems divided by rooms or air ducts, and solve them in parallel on the edge device through distributed computing to generate local optimized rotational speed sequences; summarize the local optimization results in the cloud to generate the optimized rotational speed sequence.

[0028] Furthermore, the process of dynamically adjusting the LSTM network parameters and DQN strategy by comparing the deviation between the closed-loop control prediction state and the actual environmental state, and detecting anomalies using the isolation forest algorithm includes:

[0029] Input the optimized rotational speed sequence into the dynamic thermal balance model to generate the closed-loop control prediction state in time series format;

[0030] Compare the closed-loop control prediction state with the actual environmental state, calculate the weighted Euclidean distance as the deviation, and determine whether to trigger model update;

[0031] According to the deviation results, the LSTM network parameters and the reward function weights of the DQN strategy are dynamically adjusted in the cloud, the parameters are updated and sent to the edge devices;

[0032] The Isolation Forest algorithm is used to analyze the collected data of the Internet of Things sensor network, detect abnormal states, and trigger the standby PID control strategy if abnormalities are found.

[0033] Furthermore, it is implemented through a cloud-edge collaborative architecture. The edge devices perform data processing and real-time control, and the cloud is responsible for model training and control optimization.

[0034] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0035] 1. The present invention proposes a method for generating a control reference trajectory; this method combines building thermodynamics (heat balance model), artificial intelligence (LSTM), and the Internet of Things (MQTT data transmission). The heat balance model provides physical constraints, and LSTM realizes forward-looking environmental prediction to generate a control reference trajectory; the control reference trajectory uses the predicted environmental state to guide the control and adjustment of the fan speed, thereby improving the effectiveness of the adaptive speed control of the fan for green buildings.

[0036] 2. The present invention proposes a long-term adaptive control method; this method uses a Deep Q-Network (DQN) to construct an adaptive control strategy, takes a high-dimensional feature vector and a control reference trajectory as inputs, and uses the energy saving rate, PMV thermal comfort index, and equipment life as multi-objective reward functions to generate an initial speed control sequence; DQN automatically adjusts the control strategy through online learning, reduces the dependence on manual parameter tuning, and can handle complex non-linear environments at the same time. By optimizing the multi-objective balance through the reward function, intelligent control is achieved; this method can improve the effectiveness of the adaptive speed control of the fan for green buildings.

[0037] 3. The present invention proposes a short-term control optimization method; this method constructs a Model Predictive Control (MPC) framework, combines the LSTM prediction trajectory and the DQN initial speed sequence, and optimizes the speed sequence through distributed computing; MPC dynamically optimizes the speed according to the real-time environmental state and the prediction trajectory to adapt to short-term environmental changes; distributed computing decomposes the optimization problem into sub-problems to meet the real-time control requirements of green buildings; this method can effectively improve the control accuracy, thereby improving the effectiveness of the adaptive speed control of the fan for green buildings. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a schematic flow chart of an adaptive speed control method for a fan in a green building according to the present invention;

[0039] Figure 2 It is a schematic flow chart of generating a control reference trajectory according to the present invention;

[0040] Figure 3 Flow schematic diagram for generating an optimized rotational speed sequence for the present invention. Detailed implementation manners

[0041] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part rather than all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0042] Please refer to Figures 1 to 3 , the present invention provides a method for adaptive speed control of a fan for green buildings, and the technical solution is as follows:

[0043] Embodiment 1:

[0044] In order to achieve effective adaptive speed control of a fan for a green building, a certain company used a method for adaptive speed control of a fan for a green building proposed by the present invention. The flow schematic of this method is as Figure 1 shown, and specifically includes:

[0045] Collect indoor and outdoor environmental data and fan operation status data through the Internet of Things sensor network, combine building information model data, and use the Kalman filtering algorithm for multi-source heterogeneous data fusion to generate a high-dimensional feature vector;

[0046] Further, the indoor and outdoor environmental data includes: temperature, humidity, carbon dioxide (CO2) concentration, and personnel density; the fan operation status data includes: fan speed, fan power, fan vibration, and fan temperature; the building information model data includes geometric information, material property information, and functional information;

[0047] The following gives reference fan operation status data, as shown in Table 1.

[0048] Table 1. Reference fan operation status data

[0049] Data type Reference value Fan speed 1180 rpm Fan power 295W Fan vibration <![CDATA[0.19m / s 2 > Fan temperature 39.7℃

[0050] Further, the geometric information includes the volume of the house, the cross-sectional area of the air duct, the length of the air duct, and the window area; the material property information includes thermal conductivity and heat capacity;

[0051] Further, the multi-source heterogeneous data fusion process includes: applying the extended Kalman filter algorithm to indoor and outdoor environmental data and fan operation status data, defining the state vector as [temperature, humidity, CO2 concentration, occupancy density, wind speed, fan speed, power], eliminating noise through the state transition matrix and the observation matrix, and outputting a smoothed data stream; splicing the features of the filtered data with the geometric information in the building information model data to generate a high-dimensional feature vector.

[0052] By using indoor and outdoor environmental data, fan operation status data, and building information model data, and performing multi-source heterogeneous data fusion, it is possible to provide reliable inputs for subsequent adaptive control and model predictive control, thereby improving the effectiveness of the adaptive speed control of the fan for green buildings.

[0053] Establish an indoor environmental dynamic thermal balance model, use the LSTM network to process the high-dimensional feature vector, predict the future environmental state, and generate a control reference trajectory;

[0054] Further, the process of establishing an indoor environmental dynamic thermal balance model, using the LSTM network to process the high-dimensional feature vector, predicting the future environmental state, and generating a control reference trajectory can refer to Figure 2 , specifically as follows:

[0055] Based on the principles of building thermodynamics, combined with the room volume, material thermal conductivity, and duct layout in the building information model data, establish an indoor environmental dynamic thermal balance model for calculating the changes in real-time temperature, humidity, and CO2 concentration;

[0056] Input the high-dimensional feature vector into the LSTM network, process the feature sequence within a preset time window, and obtain the predicted temperature, humidity, and carbon dioxide concentration;

[0057] Use the indoor environmental dynamic thermal balance model to impose physical constraints on the LSTM prediction results, correct the prediction deviation, and generate a control reference trajectory in the format of a time series;

[0058] Transmit the control reference trajectory from the cloud back to the edge device through the MQTT protocol as the input for adaptive control and model predictive control;

[0059] Further, the indoor environmental dynamic thermal balance model includes: a temperature thermal balance model, a humidity balance model, and a CO2 concentration balance model, and these models are all mathematical models based on equations;

[0060] Furthermore, the temperature thermal equilibrium model can be expressed as: the change rate of the indoor temperature is equal to the sum of the ventilation heat flow, the heat dissipation of people, the outdoor heat input, and the heat loss of the wall, divided by the product of the heat capacity of the air and the room volume; the humidity balance model can be expressed as: the change rate of the indoor relative humidity is equal to the moisture input brought by ventilation minus the moisture loss plus the moisture generated by people's breathing, divided by the room volume; the CO2 concentration balance model can be expressed as: the change rate of the indoor CO2 concentration is equal to the CO2 generated by people's breathing minus the CO2 removed by ventilation, divided by the room volume.

[0061] Furthermore, the LSTM network structure includes: 3 LSTM layers for predicting the future state of the indoor environment, with 128 units in each layer. The hidden layer uses the tanh activation function. A Dropout layer is added after each LSTM layer to enhance the generalization ability of the model. The fully connected layer is used to map the output of the LSTM layer to the predicted value; the preset time window is set to 30 minutes.

[0062] Furthermore, the LSTM prediction results are input into the indoor environment dynamic thermal equilibrium model to check whether the energy and mass conservation are satisfied; if it exceeds the physical range of the model, the linear interpolation method is used to correct the predicted value.

[0063] This embodiment combines building thermodynamics (thermal equilibrium model), artificial intelligence (LSTM), and the Internet of Things (MQTT data transmission). The thermal equilibrium model provides physical constraints, and LSTM realizes forward-looking environment prediction to generate a control reference trajectory; the control reference trajectory uses the predicted state of the environment to guide the adjustment of the fan speed control, thereby improving the effectiveness of the adaptive speed control of the fan for green buildings.

[0064] An adaptive control strategy is constructed using the DQN network, with the high-dimensional feature vector and the control reference trajectory as inputs, and the energy saving rate, the PMV thermal comfort index, and the equipment life as the multi-objective reward function to generate the initial speed control sequence.

[0065] Furthermore, the process of constructing an adaptive control strategy using the DQN network, with the high-dimensional feature vector and the control reference trajectory as inputs, and the energy saving rate, the PMV thermal comfort index, and the equipment life as the multi-objective reward function to generate the initial speed control sequence includes:

[0066] Construct a DQN network, define the discrete action space of the fan speed, and use the preprocessed high-dimensional feature vector and the control reference trajectory as inputs.

[0067] Design a multi-objective reward function, with the energy saving rate, the PMV thermal comfort index, and the equipment life as the optimization objectives, calculate the reward value, and guide the DQN to learn the optimal control strategy.

[0068] Train the DQN using a greedy strategy in the cloud, store state-action-reward data using the experience replay mechanism, and update the neural network parameters;

[0069] Execute DQN inference on the edge device, and generate an initial rotational speed control sequence based on the real-time input high-dimensional feature vector and the control reference trajectory;

[0070] Furthermore, the DQN consists of a 4-layer fully connected neural network with 128 neurons in each layer, and the ReLU activation function is used to enhance the nonlinear fitting ability of the model; the discrete action space of the fan rotational speed is divided according to the rotational speed range and the step size. For example, for the fan rotational speed range of 0 - 1800 rpm, the step size is set to 50 rpm, and there are 37 actions in total;

[0071] Furthermore, the high-dimensional feature vector and the control reference trajectory will be normalized before input to improve the model calculation efficiency; the high-dimensional feature vector provides real-time input for the DQN; the control reference trajectory provides the target state for the DQN and guides the calculation of the reward function;

[0072] Furthermore, the multi-objective reward function is expressed as the weighted sum of the energy-saving reward, the thermal comfort reward, the equipment life reward, and the trajectory tracking reward; the energy-saving reward represents 1 minus the ratio of the current fan power to the maximum fan power; the thermal comfort reward represents 1 minus the ratio of the absolute value of the current PMV thermal comfort index to the maximum absolute value of PMV; the equipment life reward represents 1 minus the ratio of the current fan vibration intensity to the maximum vibration intensity; the trajectory tracking reward represents 1 minus the ratio of the weighted Euclidean distance between the actual state and the control reference trajectory to the maximum distance threshold;

[0073] Furthermore, the PMV thermal comfort index is calculated through the thermal comfort model formula in the ISO 7730 standard; according to the ISO 7730 standard, the maximum absolute value of PMV is generally set to 3.

[0074] Automatically adjust the control strategy through DQN online learning, reduce the dependence on manual parameter tuning, and at the same time be able to handle complex nonlinear environments, optimize the multi-objective balance through the reward function, so as to achieve intelligent control; this can improve the effectiveness of the adaptive speed control of the fan for green buildings.

[0075] Construct a model predictive control framework, combine the control reference trajectory and the initial rotational speed control sequence, and optimize the rotational speed control sequence through distributed computing to generate an optimized rotational speed sequence;

[0076] Furthermore, the process of constructing a model predictive control framework, combining the control reference trajectory and the initial rotational speed control sequence, and optimizing the rotational speed control sequence through distributed computing to generate an optimized rotational speed sequence is as Figure 3 shown, specifically including:

[0077] Build a model predictive control framework, combine the control reference trajectory and the dynamic thermal balance model, and define the prediction horizon, control horizon, and state transition function;

[0078] Taking the initial speed control sequence as the initial value, design the optimization objective function, comprehensively consider the energy-saving rate and the PMV thermal comfort index, and set the constraint conditions;

[0079] Decompose the optimization problem into sub-problems divided by rooms or air ducts, and solve them in parallel on edge devices through distributed computing to generate a local optimized speed sequence; summarize the local optimization results in the cloud to generate an optimized speed sequence;

[0080] Furthermore, the state transition function of the model predictive control framework adopts the function formulas corresponding to the indoor environment dynamic thermal balance model, namely the temperature thermal balance function, humidity balance function, and CO2 concentration balance function; the state transition function uses the control reference trajectory as the target state; the prediction horizon is 30 minutes, and the control horizon is 5 minutes;

[0081] Furthermore, the optimization objective function of the model predictive control framework calculates the total objective value for the number of steps in the prediction horizon. The optimization objective function calculates the product of the energy-saving penalty weight and the square of the current time fan power, and the product of the thermal comfort penalty weight and the square of the deviation between the current time PMV thermal comfort index and the target PMV, and then sums them up; the number of steps in the prediction horizon is the ratio of the prediction horizon to the step size, and the step size is 5 seconds;

[0082] Furthermore, the constraint conditions are speed range constraints, PMV range constraints, and power range constraints.

[0083] Respectively give the initial speed values and optimized speed values of the initial speed control sequence and the optimized speed sequence at different time steps. The initial speed values are obtained by DQN, and the optimized speed values are obtained by the model predictive control framework (MPC); the speed comparison results are shown in Table 2.

[0084] Table 2. Speed Comparison Results

[0085] Time step Initial speed value Optimized speed value 300s 1250 rpm 1230 rpm 900s 1300 rpm 1270 rpm 1500s 1290 rpm 1280 rpm

[0086] By constructing a model predictive control framework, combining the LSTM prediction trajectory and the DQN initial speed sequence, and using distributed computing to optimize the speed sequence; MPC dynamically optimizes the speed according to the real-time environmental state and the prediction trajectory to adapt to short-term environmental changes; distributed computing decomposes the optimization problem into sub-problems to meet the real-time control requirements of green buildings; this method can effectively improve the control accuracy, thereby improving the effectiveness of the adaptive speed control of fans for green buildings.

[0087] Input the optimized rotational speed sequence into the dynamic thermal balance model to generate the predicted state of closed-loop control; by comparing the deviation between the predicted state of closed-loop control and the actual environmental state, dynamically adjust the LSTM network parameters and DQN strategy, and use the Isolation Forest algorithm to detect anomalies.

[0088] Further, the process of dynamically adjusting the LSTM network parameters and DQN strategy and using the Isolation Forest algorithm to detect anomalies by comparing the deviation between the predicted state of closed-loop control and the actual environmental state includes:

[0089] Input the optimized rotational speed sequence into the dynamic thermal balance model to generate the predicted state of closed-loop control in the format of a time series;

[0090] Compare the predicted state of closed-loop control with the actual environmental state, calculate the weighted Euclidean distance as the deviation, and determine whether to trigger model update;

[0091] According to the deviation result, dynamically adjust the reward function weights of the LSTM network parameters and DQN strategy in the cloud, update the parameters and send them down to the edge device;

[0092] Use the Isolation Forest algorithm to analyze the collected data of the Internet of Things sensor network, detect the abnormal state, and trigger the standby PID control strategy if an anomaly is found;

[0093] Further, input the optimized rotational speed sequence into the dynamic thermal balance model to update the ventilation heat flow term, which is related to the fan rotational speed, solve it on the edge device using the Runge-Kutta 4th order method, with a prediction time domain of 5 minutes, and generate the predicted state of closed-loop control;

[0094] Further, compare the corresponding temperature, humidity, and CO2 concentration in the predicted state of closed-loop control and the actual environmental state; the deviation threshold is set to 0.8;

[0095] Further, if the deviation exceeds the threshold, the cloud uses the PyTorch framework to perform online fine-tuning on the parameters of the LSTM model. The fine-tuning is executed every 5 minutes, update the parameters (weights and biases) and send them down to the edge device; at the same time, perform fine-tuning on the DQN reward function weights in the cloud, update the parameters and send them to the edge device through MQTT;

[0096] Further, select temperature, CO2 concentration, fan vibration, and deviation as the main analysis objects of the Isolation Forest algorithm, with the number of trees being 100, the proportion of anomaly points being 0.05, and the maximum number of samples being 256; calculate the average path length of each data point in 100 trees to obtain the anomaly score; if the anomaly score exceeds 0.6 or the key feature fluctuation exceeds the limit, it is determined as an anomaly.

[0097] By inputting the optimized speed sequence into the heat balance model to generate a closed-loop prediction state, comparing the deviation to adjust the LSTM and DQN parameters, and using isolation forest to detect anomalies, the real-time control effect can be accurately evaluated, thereby assisting the effectiveness of the adaptive speed control of fans for high green buildings.

[0098] The method used in this embodiment is implemented through a cloud-edge collaborative architecture, where the edge device performs data processing and real-time control, and the cloud is responsible for model training and control optimization. This architecture can be used to achieve cross-domain technology integration, dynamic adaptation and learning capabilities, and improvements in real-time and response speed, thereby improving the effectiveness of adaptive speed control of fans for green buildings.

[0099] This embodiment proposes an adaptive speed control method for a fan for a green building. First, indoor and outdoor environment and operation data are collected, and multi-source heterogeneous data are fused to generate a high-dimensional feature vector in combination with building information model data. A dynamic thermal balance model is established, and an LSTM network is used to predict the future environmental state and generate a control reference trajectory. A DQN network is used to construct an adaptive control strategy, and an initial speed control sequence is generated with the high-dimensional feature vector and the control reference trajectory as input. A model predictive control framework is constructed, and the speed control sequence is optimized through distributed computing in combination with the control reference trajectory and the initial speed control sequence to generate an optimized speed sequence. By comparing the deviation between the closed-loop control predicted state and the actual state, the LSTM network parameters and the DQN strategy are adjusted and anomalies are detected. The present invention can improve the effectiveness of adaptive control in an Internet of Things control system.

[0100] Embodiment 2:

[0101] The present invention proposes an adaptive speed control method for a green building fan. In order to further verify the effectiveness of the method flow proposed by the present invention, this embodiment applies the method in a green building (a 1,000 square meter office building equipped with a variable frequency centrifugal fan with a rated power of 500 watts and a maximum speed of 1,800 revolutions per minute) to conduct a method flow effectiveness test.

[0102] The present invention selects historical data of a green building ventilation system in the past three years as the data set of the method flow, takes the data from the 1st to 2nd year as the training set, and takes the data from the 3rd year as the verification set; the sampling rule is: take the week as the unit, and randomly select 5 days of data per week; among them, 2 groups of data are collected every day in the morning (8:00-10:00), noon (12:00-14:00) and evening (17:00-19:00), and each group lasts for 30 minutes; the data covers the same type of fan (centrifugal fan, rated power 500 watts) to ensure consistency.

[0103] Four different control schemes are respectively used to process the same historical test data to obtain the closed-loop control prediction states of each scheme. Among them, the historical test data includes historical indoor and outdoor environmental data and historical fan operation state data, and the building information model data can be directly obtained by calling. Then, the weighted Euclidean distance between the closed-loop control prediction state and the actual environmental state is calculated to obtain the deviation of each scheme, and it is compared with the deviation threshold to obtain the proportion of the control results of each scheme within a reasonable range.

[0104] The control schemes are respectively: the adaptive speed control scheme proposed by the present invention, that is, "control reference trajectory generation + DQN adaptive control + model predictive control", denoted as Scheme 1; removing the control reference trajectory generation process, denoted as Scheme 2; removing the DQN adaptive control process, denoted as Scheme 3; removing the model predictive control process, denoted as Scheme 4.

[0105] The test results of the effectiveness of the control schemes are shown in Table 3.

[0106] Table 3 Test results of the effectiveness of the control schemes

[0107]

[0108] It can be seen from the results in Table 3 that the control scheme proposed by the present invention has better test results than those of other schemes in the effectiveness test results. Thus, it can be shown that the scheme of "control reference trajectory generation + DQN adaptive control + model predictive control" proposed by the present invention can obtain relatively accurate control results, thereby improving the effectiveness of the adaptive speed control of the fan for green buildings.

[0109] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirits of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An adaptive speed control method for a fan used in a green building, characterized in that, Including: Collect indoor and outdoor environmental data and operating status data through the Internet of Things sensor network, combine with building information model data, and use the Kalman filter algorithm for multi-source heterogeneous data fusion to generate high-dimensional feature vectors; Establish an indoor environmental dynamic thermal balance model, use the LSTM network to process the high-dimensional feature vectors, predict the future environmental state, and generate a control reference trajectory; Adopt a DQN network to construct an adaptive control strategy, use the high-dimensional feature vectors and the control reference trajectory as inputs, and use the energy saving rate, PMV thermal comfort index, and equipment life as multi-objective reward functions to generate an initial rotational speed control sequence; Construct a model predictive control framework, combine the control reference trajectory and the initial rotational speed control sequence, and optimize the rotational speed control sequence through distributed computing to generate an optimized rotational speed sequence; Input the optimized rotational speed sequence into the dynamic thermal balance model to generate a closed-loop control prediction state; by comparing the deviation between the closed-loop control prediction state and the actual environmental state, dynamically adjust the LSTM network parameters and DQN strategy, and use the Isolation Forest algorithm to detect anomalies.

2. The adaptive speed control method for a fan used in a green building according to claim 1, wherein, The indoor and outdoor environmental data includes: temperature, humidity, carbon dioxide concentration, and personnel density; the operating status data includes: rotational speed, power, vibration, and temperature; the building information model data includes geometric information, material property information, and functional information.

3. The adaptive speed regulation control method for a fan used in a green building according to claim 1, characterized in that, The generation process of the control reference trajectory includes: Based on the building information thermodynamics principle, combine with the room volume, material thermal conductivity, and duct layout in the building information model data to establish an indoor environmental dynamic thermal balance model for calculating the changes in real-time temperature, humidity, and CO2 concentration; Input the high-dimensional feature vectors into the LSTM network, process the feature sequence of a preset time window, and obtain the predicted temperature, humidity, and carbon dioxide concentration; Use the indoor environmental dynamic thermal balance model to impose physical constraints on the LSTM prediction results, correct the prediction deviation, and generate a control reference trajectory in the format of a time series; Transmit the control reference trajectory from the cloud back to the edge device through the MQTT protocol as the input for adaptive control and model predictive control.

4. The adaptive speed regulation control method for a fan used in a green building according to claim 1, wherein The process of adopting a DQN network to construct an adaptive control strategy and generating an initial rotational speed control sequence includes: Construct a DQN network, define a discrete action space, and use the preprocessed high-dimensional feature vectors and the control reference trajectory as inputs; Design a multi-objective reward function, define the optimization objective, calculate the reward value, and guide the DQN network to learn the optimal control strategy; Train the DQN network using the greedy strategy in the cloud, use the experience replay mechanism to store state-action-reward data, and update the neural network parameters; Execute DQN inference on the edge device, and generate the initial rotational speed control sequence according to the real-time input high-dimensional feature vectors and the control reference trajectory.

5. The adaptive speed regulation control method for a fan used in a green building according to claim 1, wherein, The process of constructing a model predictive control framework, combining the control reference trajectory and the initial rotational speed control sequence, and optimizing the rotational speed control sequence through distributed computing to generate an optimized rotational speed sequence includes: Construct a model predictive control framework, combine the control reference trajectory and the dynamic thermal balance model, and define the prediction horizon, control horizon, and state transition function; Taking the initial speed control sequence as the initial value, design an optimization objective function, comprehensively consider the energy-saving rate and the PMV thermal comfort index, and set the constraint conditions; Decompose the optimization problem into sub-problems divided by rooms or air ducts, and solve them in parallel on edge devices through distributed computing to generate a local optimized speed sequence; summarize the local optimization results in the cloud to generate an optimized speed sequence.

6. The adaptive speed regulation control method for a fan used in a green building according to claim 5, characterized in that, The state transition function adopts the same formula as the indoor environment dynamic thermal balance model.

7. A method for adaptive speed regulation control of a fan for green buildings according to claim 1, characterized in that, The process of dynamically adjusting the LSTM network parameters and DQN strategy and detecting anomalies using the isolation forest algorithm by comparing the deviation between the predicted state of the closed-loop control and the actual environmental state includes: Input the optimized speed sequence into the dynamic thermal balance model to generate the predicted state of the closed-loop control in the format of a time series; Compare the predicted state of the closed-loop control with the actual environmental state, calculate the weighted Euclidean distance as the deviation, and determine whether to trigger model update; According to the deviation result, dynamically adjust the reward function weights of the LSTM network parameters and DQN strategy in the cloud, update the parameters and send them to the edge device; Use the isolation forest algorithm to analyze the collected data of the Internet of Things sensor network, detect abnormal states, and trigger the standby PID control strategy if anomalies are found.

Images